Exploring individual variation in constructional schematicity using random effects

Usage-based constructionist approaches see language as an inventory of constructions at different levels of schematicity learned from the input. If so, personal construct-i-cons should vary as a function of usage. Repeated use and chunking/entrenchment of specific instances should lead to reanalysis of their internal structure and change in the level of schematicity. Emancipation of such instances can be viewed as idiosyncratic node creation in the network of personal constructions. This paper exploits the variation between the full and the reduced form of it is to detect instances undergoing reanalysis in a 1.75-million-word diachronic corpus of a single blogger. Occurrence of the reduced form is modelled as an effect of time, constructional entropy and unidirectional word-to-construction and construction-to-word associations. Including lexically specified instances of different constructions as random effects reveals variation in schematicity (intercept variance) and change in schematicity over time (slope variance for the effect of temporal order).